Targeted activation of spinal respiratory neural circuits

Abstract

Spinal interneurons which discharge in phase with the respiratory cycle have been repeatedly described over the last 50 years. These spinal respiratory interneurons are part of a complex propriospinal network that is synaptically coupled with respiratory motoneurons. This article summarizes current knowledge regarding spinal respiratory interneurons and emphasizes chemical, electrical and physiological methods for activating spinal respiratory neural circuits. Collectively, the work reviewed here shows that activating spinal interneurons can have a powerful impact on spinal respiratory motor output, and can even drive rhythmic bursting in respiratory motoneuron pools under certain conditions. We propose that the primary functions of spinal respiratory neurons include 1) shaping the respiratory pattern into the final efferent motor output from the spinal respiratory nerves; 2) coordinating respiratory muscle activation across the spinal neuraxis; 3) coordinating postural, locomotor and respiratory movements, and 4) enabling plasticity of respiratory motor output in health and disease.

1. Overview of the spinal respiratory circuit

Centuries of work unequivocally establish that the brainstem set the fundamental respiratory rhythm in mammals; however, spinal cord interneuron networks also play a role in regulating the activity of spinal respiratory motoneurons. A spinal interneuronal substrate for transmitting bulbospinal respiratory input to the contralateral side of the spinal cord was suggested in 1946 (Pitts, 1946), and a multisynaptic reflex pathway was suggested a few years later for the intercostal motor circuit (Downman, 1955). The laboratories of Kirkwood, Duffin, Aoki and others have focused on spinal respiratory interneurons over the last 30 years. As one example, in 1986, Lipski and Duffin concluded that, “cervical neurons are involved in the control of phrenic and intercostal motoneurons, probably through a disynaptic pathway involving segmental interneurons” (Lipski and Duffin, 1986). In recent years interest in spinal interneurons and breathing has been renewed, particularly in the context of motor recovery after spinal cord injury (Lane et al., 2008).

The spinal respiratory circuitry includes a diffuse distribution of interneurons in the cervical and thoracic spinal cord (Fig. 1). These interneurons have synaptic connections with spinal respiratory motoneurons and also appear to be synaptically coupled with central pattern generators (CPGs) in the cervical and lumbar spinal cord. In addition, a rhythmically active group of interneurons in the C1–C2 spinal cord has been identified (Aoki et al., 1980; Aoki et al., 1978; Kobayashi et al., 2010). We emphasize that formal classification of propriospinal neurons in a particular functional category (i.e., “respiratory”, “locomotor”, etc.) is likely to be complicated by the multifunctional nature of spinal interneurons (Jankowska, 2001). Thus, bursting patterns and/or functional connectivity of propriospinal neurons will almost certainly depend on the particular experimental conditions. For purposes of this review, we define the spinal respiratory circuit to include propriospinal or segmental interneurons neurons that burst in phase with the respiratory cycle (i.e. inspiratory or expiratory) and/or are synaptically coupled to brainstem premotor respiratory neurons or to spinal respiratory motoneurons. In this article, we review some of the key historical data describing respiratory interneurons in the spinal cord, and provide an overview of physiological, chemical and electrical methods which can activate spinal respiratory networks. The reader is also referred to prior reviews, which have covered aspects of this topic (Hormigo et al., 2017; Jensen et al., 2020; Zaki Ghali et al., 2019; Zholudeva et al., 2018).

Fig. 1.

Spinal interneuron connectivity. Respiratory rhythm generated in the brainstem is transmitted to spinal neurons via bulbospinal premotor neurons (axons not shown, but project to all spinal populations). Upper cervical interneurons (UCINs, blue) reside within C1–C2 and may be capable of generating a rhythm. Cervical (purple) and lumbar (green) locomotor central pattern generators (CPGs) can produce rhythmic patterns recorded on respiratory nerves. Respiratory interneurons (pink) modulate bulbospinal rhythm and coordinate activity between motor pools. Interneurons can task switch and it is likely that interneuron classification is not static (cells in both networks). Respiratory (pink) and limb (purple and green) afferents project towards spinal motor neurons and can modify respiratory pattern generation. Example inspiratory (left) and expiratory (right) motor pools are shown. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

1.1. Anatomy of spinal respiratory circuitry

The final respiratory motor output will occur via alpha motoneurons, and the distribution of the spinal motoneurons which innervate the respiratory muscles is summarized in Fig. 2. Innervation of the accessory sternocleidomastoid and trapezius muscles forms at C1–C2, the diaphragm at C3–C5/6, scalenes at C4–C8, and external intercostal T1–T11. Expiratory muscle innervation similarly spans the cord form C5–L3 beginning with the pectoralis major at C5–C8, internal intercostal at C8–T11, and several abdominal muscles from T6–L3. While the location of motor neuron pools can be mapped using standard anatomical tracing methods (as shown by the histological images in Fig. 2), the diffuse distribution of the spinal interneurons which have the ability to modulate respiratory motor output is more challenging to determine. Transynaptic viral tracers provide a means of studying the spinal respiratory circuit, and have been used to confirm synaptic connectivity between spinal interneurons and respiratory motor pools (Lane et al., 2009). Determining mono- vs. polysynaptic connectivity with transynaptic tracers is challenging, but by carefully titrating the post-inoculation time window, Lane et al. provided compelling evidence for propriospinal neurons that are monosynaptically coupled to phrenic motoneurons (Lane et al., 2009). These “pre-phrenic interneurons” (Zaki Ghali et al., 2019) were found throughout the cervical cord including the dorsal horn, laminae VII, laminae X (near the central canal). Additional evidence for “pre-phrenic interneurons” was provided using viral tracers in the rat (Dobbins and Feldman, 1994), mouse (Qiu et al., 2010), cat (Lois et al., 2009) and ferret (Yates et al., 1999), although additional work is needed to unequivocally confirm these observations.

Fig. 2.

Anatomy of spinal respiratory circuit. Locations of motor nerves for expiratory (top) and inspiratory (bottom) muscles are denoted by colored bars. Example histological sections showing cross sectional location of motor neurons from muscles traced with wheat germ agglutinin (callouts).

The location of interneurons associated with non-phrenic respiratory muscles is less well described. Early work by Eccles suggested interneurons in the thoracic spinal cord are part of spinal reflexes that alter expiratory motoneuron output (Eccles et al., 1962). Subsequently, interneurons in the thoracolumbar spinal cord were shown to be synaptically coupled to abdominal motoneurons (Billig et al., 1999; Billig et al., 2001). This work used rabies virus as a tracer, and showed that abdominal neurons typically reside ipsilateral to viral injection. The rectus and tranversus abdominus neuron distribution extended from T1–L4, though the rectus had a broad distribution peaking from T10-T13, while the transversus had a narrower distribution peaking between T12 and T13. These studies could not clearly differentiate between motor and premotor neurons and the large peaks likely indicate the location of the motor neuron pools. There was, however, a relatively flat (average ~5 cells/segment/animal) contralateral distribution of neurons from T1–L4 which can be presumed to be interneurons. While there were dual labeled medullary interneurons, the authors did not describe propriospinal interneurons that were shared between the rectus and transversus abdominals.

To our knowledge, only one anatomical study has investigated the overlap of respiratory interneurons within the spinal cord across multiple motoneuron pools (Lane et al., 2008). Pseudo rabies virus was injected into both the intercostal and diaphragm muscles, this procedure produced dual labeled (and therefore synaptically coupled) spinal interneurons. Further, there are two studies using transneuronal tracing which identified common interneurons in the brainstem (Billig et al., 2000; Shintani et al., 2003) synaptically coupling tongue (genioglossus) or abdominal motor pools to the phrenic motor pool. Overall, relatively little information is available in regards to “functional overlap” within and across spinal respiratory interneuron populations.

1.2. Connectivity between brainstem and spinal respiratory interneurons

Direct monosynaptic connectivity between brainstem respiratory centers and phrenic motoneurons is well-established (Dobbins and Feldman, 1994; Ellenberger and Feldman, 1988; Ellenberger et al., 1990). Available evidence suggests that these monosynaptic connections are primarily responsible for diaphragm inspiratory contraction, at least during standardized laboratory conditions (Cohen et al., 1974; Rosenbaum and Renshaw, 1949). However, there is also strong evidence for synaptic connectivity between brainstem respiratory neurons and spinal cord interneurons, and these connections also play a role in respiratory motor control. The first such evidence probably comes from Gesell et al. (1936). Systematic extracellular recording throughout the brainstem and upper cervical cord of dogs revealed “extensive respiratory potentials” in the spinal cord which “conform with the necessary chain of impulses for the innervation of respiratory muscles” (Gesell et al., 1936). Further, these upper cervical spinal neurons fired action potentials during both inspiration and expiration. Subsequently, many studies have reported inspiratory and expiratory discharge patterns of spinal interneurons (reviewed in Section 2.1). These electrophysiology studies are supported by anatomical studies using anterograde tracing to reveal bulbospinal projections from the ventral respiratory group which terminate in the immediate vicinity of cervical respiratory interneurons in rats (Lane et al., 2008) and mice (Qiu et al., 2010).

Synaptic connectivity between brainstem neurons and spinal cord interneurons is also indicated by several electrophysiological studies employing correlational methods to assess the temporal relationships between brainstem and spinal neuron discharge (Hoskin and Duffin, 1987a; Hoskin and Duffin, 1987b). For example, Tian and Duffin described monosynaptic connections between inspiratory bulbospinal respiratory neurons and upper cervical interneurons (Tian and Duffin, 1996b). They used cross correlational analysis of action potential spikes recorded in medullary inspiratory neurons (trigger) and upper cervical interneurons (target) to show clear, albeit sparse, synaptic connections. Cross correlation with phrenic nerve activity (target) indicated that the same bulbospinal neurons also synapsed onto phrenic motoneurons. Additional cross correlational analysis of brainstem recorded spikes with internal and external intercostal nerve activity showed long latency peaks suggesting polysynaptic pathways from bulbospinal neurons to intercostal motoneurons. This finding is consistent with work from Kirkwood’s laboratory indicating that spinal interneurons play a significant role in modulating thoracic respiratory motor output (Kirkwood, 1995; Kirkwood et al., 1988).

Collectively, the available literature indicates that synaptic connections between brainstem respiratory neurons and spinal cord interneurons are present, but considerable work remains to fully elucidate the location and functional significance of these pathways. The respiratory-related firing patterns of spinal interneurons is most likely due to these bulbospinal connections, however, other explanations are possible including spinally-generated respiratory rhythms (Aoki et al., 1980; Aoki et al., 1978; Kobayashi et al., 2010) (see Section 2.1) and/or synaptic connections with respiratory muscle and chest wall afferent neurons (Nair et al., 2017a) (see Section 3.7).

1.3. Electrophysiological evidence for direct synaptic connections between spinal interneurons and respiratory motoneurons

Conclusions regarding spinal synaptic connections can be inferred by mathematical techniques such as cross correlation and/or spike triggered averaging (STA). For example, cross correlation can be used to evaluate the temporal relationship between interneuron discharge (spikes) and compound action potentials recorded in a motor nerve (e.g. phrenic). Averaging the discharge of respiratory nerve or muscle activity based on spinal interneuron spike timing (i.e., STA) can provide insight into the connectivity between spinal interneurons and motoneurons. Using these methods, Lipski and Duffin found no evidence for monosynaptic connections between upper cervical interneurons and the phrenic or intercostal motor pools in the cat. However, when they recorded intracellularly in intercostal motoneurons, STAs revealed features suggestive of disynaptic connections with upper cervical interneurons (Lipski and Duffin, 1986). Additional studies in the cat indicated extensive projections from upper cervical neurons towards the phrenic and intercostal motor pools, with long latency features indicating polysynaptic pathways. However, those experiments did reveal STA features indicative of monosynaptic connections in the phrenic or intercostal nerves (Douse et al., 1992). In contrast, upper cervical interneurons activated by L-glutamate show STA features in the phrenic nerves with latencies suggesting monosynaptic connections (Nakazono and Aoki, 1994). Another study in rats showed that upper cervical interneuron spikes produced STA features in the ipsilateral phrenic nerve consistent with monosynaptic connections, and no such features were present in the intercostal nerves or contralateral phrenic nerve (Tian and Duffin, 1996a). However, the data also indicated that upper cervical interneurons make multisynaptic connections with phrenic motoneurons. Accordingly, this may be the most biologically conserved route of excitation between upper cervical interneurons and phrenic motoneurons since it has now been reported in both rat (Tian and Duffin, 1996a) and cat (Douse et al., 1992).

Mid-cervical (C5) expiratory interneurons in the cat show evidence for monosynaptic inhibitory connections to phrenic motor neurons (Douse and Duffin, 1993) or no connection (Duffin and Iscoe, 1996). Sandhu et al. reported that 5% of recorded cervical interneurons in the C3–C4 rat spinal cord produced a short latency (~7 ms) STA peak in the ipsilateral phrenic nerve recording, consistent with an excitatory monosynaptic connection (Sandhu et al., 2015). More recent work from our group has shown that mid cervical (C3–C5) interneurons in the rat have mono- and disynaptic connections to phrenic motoneurons based on STA of phrenic nerve activity (Streeter et al., 2017a). We also noted monosynaptic inhibitory connections of C3–C5 interneurons with the contralateral phrenic motor pool. In the feline thoracic spinal cord, commissural interneurons can produce both tonic excitation as well as phasic inhibition of intercostal motoneurons as revealed by STA of focal field potentials (Kirkwood et al., 1993). Collectively, the available electrophysiological evidence makes it clear that a spinal interneurons can have direct synaptic connections with respiratory motoneurons, but considerable work is needed to map the distribution of these cells in both health and disease. Multi-electrode arrays (Sandhu et al., 2015; Streeter et al., 2017b) and/or opto- (Cregg et al., 2017) and chemogenetic (Jensen et al., 2019; Satkunendrarajah et al., 2018) approaches (see Sections 2.1, 3.1 and 3.2) will prove highly useful in this regard.

1.4. Electrical stimulation of the spinal cord: Implications for interneuronal connectivity

Electrical stimulation of the spinal cord may have therapeutic value (see Section 3.3), and is also a useful experimental approach to elucidate synaptic connections in the spinal respiratory circuit (Kirkwood et al., 1988; McCrimmon et al., 1997). Kirkwood and colleagues stimulated within the thoracic spinal cord of cats using either a fixed tungsten array or a mobile tungsten electrode (0.2 ms, < 20 V) while concurrently recording from respiratory interneurons. Stimulation revealed that the vast majority of these interneurons projected to more caudal regions of the thoracic spinal cord with most axons projecting to the contralateral side (Kirkwood et al., 1988). This is a landmark study since it verified the presence of a dense respiratory interneuron network within the thoracic spinal cord. Subsequent work from Kirkwood’s laboratory indicates that thoracic interneurons are an important component of the respiratory-related control of the intercostal muscles (Kirkwood et al., 1993; Saywell et al., 2011).

The latency of spinal-stimulation evoked potentials recorded in nerves or muscles (e.g., stimulus to onset of evoked potentials or STA peaks) can be used to infer interneuron synaptic connections in the spinal cord (Ling et al., 1994, 1995; Sunshine et al., 2018). Ling et al. described long-latency (and therefore polysynaptic) spinal pathways to phrenic motoneurons, but these pathways were only evoked following systemic administration of a serotonergic precursor molecule (Ling et al., 1994, 1995). A recent study systematically stimulated throughout the spinal cord (from C2–T1) while recording evoked potentials in the sternomastoid, external intercostal and diaphragm muscles in rats. Long latency (2.6–3.9 ms) evoked potentials could be readily evoked in each of these muscles when the spinal cord was stimulated during the inspiratory period. This finding is consistent with the presence of polysynaptic spinal pathways to spinal respiratory motoneurons in the spinal cord. In contrast, stimulation during the expiratory phase restricted the sites which could elicit evoked potentials to those nearest the motor pool, particularly for the diaphragm (Sunshine et al., 2018). This latter finding may indicate inhibition of pre-inspiratory spinal interneurons during the expiratory phase.

Collectively, the published studies of electrical spinal stimulation indicate the presence of both mono- and polysynaptic spinal pathways to respiratory motoneurons. These complex spinal pathways and networks may or may not be directly relaying respiratory-related synaptic drive from the brainstem to spinal respiratory motoneurons (this remains an open question), but it is likely that these pathways serve to integrate information from multiple brainstem, spinal, and afferent sources to modulate respiratory output. A summary of the distribution of respiratory interneurons in the spinal cord is provided in Fig. 1, and is based on anatomical and neurophysiological studies summarized in Sections 1.1–1.4.

2. Behavior of the spinal respiratory circuit

Breathing, and thus the activity of respiratory neurons and networks, changes dynamically with metabolic demand/physical activity, posture, time of day, sleep-wake cycle, etc. In addition, the respiratory neural control network is capable of considerable plasticity on both short and longer time scales (Fuller and Mitchell, 2017). It is also well established that spinal interneuron circuits are highly adaptable to prevailing conditions (Jankowska, 2001), and in regards to spinal respiratory networks both injury and disease can lead to considerable remodeling (Zholudeva et al., 2017). Thus, we suggest that it is difficult to draw firm conclusions regarding spinal interneuron function based on the discharge patterns recorded during a particular experimental condition (Jankowska, 2001). Rather, conclusions need to be restricted to the particular condition, with the understanding that the physiologic role of the particular cell or propriospinal network may be different if circumstances are changed. Nevertheless, studies of interneuron discharge patterns and network behavior are fundamentally important to help to unravel the complexity of spinal respiratory motor control, and here we review prior work in this area.

2.1. Respiratory-related spinal interneuron firing patterns

To our knowledge the first electrophysiological studies to indicate respiratory-related discharge patterns of spinal cord interneurons were published by Sumi in 1963. Extracellular recording in the thoracic spinal cord of cats identified presumed interneurons that burst with inspiratory or expiratory firing patterns (Sumi, 1963a, 1963b). Subsequently, respiratory-related interneuron bursting has become a common finding in studies of cervical and thoracic interneuron discharge.

Interneurons in the high cervical spinal cord (C1–C2) of cats can fire rhythmically during the inspiratory phase following transection of the cord at C1 (Aoki et al., 1980). Inspiratory modulated interneurons in this region project towards thoracic and lumbar motor pools in cats (Hoskin et al., 1988), and these cells may act to inhibit expiratory motoneurons (Lipski and Duffin, 1986). Other studies confirm projections of high cervical interneurons towards intercostal motor pools (Lipski and Duffin, 1986). Propriospinal C1–C3 neurons in the cat receive phasic excitation during inspiration and inhibition during expiration (Duffin and Hoskin, 1987), and neurons from this region also project within the phrenic motor pool (Lipski and Duffin, 1986).

Several studies have evaluated mid-cervical (C3–C5) interneuron discharge in relation to the respiratory cycle (Bellingham and Lipski, 1990; Duffin and Iscoe, 1996; Palisses et al., 1989). Palisses et al. recorded respiratory interneurons in the rabbit, and concluded that they were part of the circuit responsible for activating phrenic motoneurons (Palisses et al., 1989). Subsequently, Bellingham and Lipski reported that C5 interneurons in the cat have strong respiratory modulation with 42% of recorded cells showing inspiratory modulation and 52% showing expiratory modulation, the remaining cells were classified as post-inspiratory (Bellingham and Lipski, 1990). Streeter et al. used a multielectrode array to record rat spinal interneurons and found that at baseline the majority (76%) of C4–C5 interneurons had a tonic firing pattern with no apparent respiratory modulation, while 17% burst primary during inspiration and 7% during expiration. Interestingly, acute respiratory stimulation with hypoxia activated a population of expiratory modulated cells which had been quiescent during baseline recordings. Moreover, repeated bouts of hypoxia increased the number of detectable mid-cervical interneurons which produced STA features in the phrenic neurograms, suggesting that the intermittent hypoxia exposure resulted in an increase in the synaptic connectivity between interneurons and phrenic motoneurons (Streeter et al., 2017a).

The thoracic respiratory interneuron network has been extensively investigated by Kirkwood and colleagues (Enriquez Denton et al., 2012; Kirkwood, 1995; Kirkwood et al., 1988; Kirkwood et al., 1993; Saywell et al., 2011; Schmid et al., 1993). An initial study described inspiratory and expiratory discharge patterns of thoracic interneurons, and the authors suggested that there were approximately 10-fold more respiratory thoracic interneurons as compared to bulbospinal neurons (Kirkwood et al., 1988). Subsequent work showed that intercostal motoneurons receive synaptic input from contralateral interneurons (Schmid et al., 1993), and that intercostal motoneurons receive tonic excitatory interneuron input layered with phasic inhibitory interneuron input (Kirkwood et al., 1993). In that study, the majority of the recorded interneurons were inhibitory and fired throughout the respiratory cycle, though there were neurons which increased their firing during either the inspiratory or expiratory phase.

While differences exist across experimental conditions and species (e.g., the proportion of cells showing a particular discharge pattern, recording locations, etc.), the common finding across all electrophysiological investigations of mammalian spinal interneurons is that some proportion of the recorded cells discharge in phase with the respiratory cycle (either inspiration or expiration). Thus, information regarding breathing (e.g., bulbospinal neural signals, chest wall movements, lung inflation, etc.) is being synaptically relayed to these cells. The possibility that spinal interneurons may, in some cases, be responsible for generating a respiratory rhythm is considered next.

2.2. Rhythmogenic spinal interneuron circuits: Implications for breathing

Spinal interneurons are capable of generating rhythmic discharges in respiratory motoneuron pools. The concept of spinal respiratory rhythm generation was summarized by Feldman in his excellent 1986 review article on the neural control of breathing (Feldman, 1986). The fundamental observation, across experiments from multiple laboratories, is as follows, after complete spinal cord section rostral to the phrenic motor nucleus, rhythmic motor output can be evoked in the diaphragm or phrenic nerve. However, this response is typically not observed “spontaneously”, but rather requires specific conditions that are created by spinal application of drugs and/or afferent input from proprioceptors. One exception was provided by Zaki-Ghali and Marchenko, who described spontaneous appearance of phasic bursting in the phrenic nerves of Sprague-Dawley adult rats after transection of the spinal cord at C1–C2 (this response occurred in 7 of 20 experiments). However, they also showed that pharmacological blockade of inhibitory synaptic transmission in the cervical spinal cord produced a much more robust phrenic rhythm. This manifested as large, rhythmic bursts in the left and right phrenic nerves that had a decrementing pattern within each burst (Ghali and Marchenko, 2016).

The high cervical spinal cord is capable of generating rhythmic output in the phrenic neve (Aoki et al., 1978; Coglianese et al., 1977). Aoki and colleagues have described respiratory-related interneurons in C1–C2 (Aoki et al., 1980), which they have termed “upper cervical inspiratory neurons” (UCINs). This group of neurons is innervated by brainstem respiratory neurons (Aoki et al., 1988), and has direct and also polysynaptic connections to phrenic motoneurons in the mid-cervical spinal cord (Nakazono and Aoki, 1994; Tian and Duffin, 1996a). The UCIN group appears to play a role in shaping the pattern of respiratory motor output. In addition, work in the spinal cord slice preparation suggests that UCINs are capable of spontaneously generating a rhythm (Kobayashi et al., 2010). This is consistent with the initial report in 1978 from Aoki et al. showing rhythmic phrenic bursting in spinal C1 (Aoki et al., 1978), but not C3 (Aoki et al., 1980) transected cats. Viala and Frenton suggested that the mid-cervical cord contains a respiratory rhythm generator. Using a curarized rabbit preparation where respiratory rhythms stopped following C2–C3 transection, but were restored following application of nialamide (monoamine oxidase inhibitor) and DOPA which likely acts on monoamernergic terminals to release catecholamines (Viala and Freton, 1983). A subsequent study used progressive sectioning in neonatal rats to localize this “spinal respiratory generator” to C4–C6 (Dubayle and Viala, 1996). However, we suggest that it is unlikely that this circuit is a CPG dedicated to “respiratory rhythms” per se, but rather is a CPG that is relevant for upper extremity locomotor movements that can, under some circumstances, drive a rhythm in the phrenic nerves. For example, work from Morin’s laboratory shows that propriospinal connections from locomotor pattern generators in the lumbar spinal cord can produce rhythmic activity in phrenic motor pools (Le Gal et al., 2016; Le Gal et al., 2014). This work built upon earlier reports from Viala and colleagues (Viala, 1986; Viala et al., 1987; Viala et al., 1979), and is further supported by Schomburg et al. (Schomburg et al., 2003). Thus, locomotor activity is one condition that is capable of driving rhythmic spinal respiratory activity, and this may play a role in coordinating breathing with locomotion (see Section 3.6).

In recent years, studies of spinal respiratory networks, particularly in regards to pattern generators, has been advanced using optoand chemogenetic approaches. Alilain et al. first reported that a viral construct driving a rhodopsin could be used to enable light activation of spinal neurons and activation of the diaphragm (Alilain et al., 2008). Diaphragm activation appeared to involve interneuron stimulation, although direct evidence was lacking. However, rhythmic respiratory activity below the site of a spinal injury (C2 hemilesion) persisted for up to an hour after the light activation was halted, suggesting that the initial stimulation of spinal circuits was able to activate a self-sustaining rhythmicity that could maintain breathing for a significant period of time. Subsequently, Cregg et al. described a cervical spinal network that was capable of rhythmically activating phrenic motoneurons in the absence of brainstem inputs (Cregg et al., 2017). The rhythmic bursting could be evoked by pharmacologic blockade of synaptic inhibition in the cervical spinal cord via application of strychnine and picrotoxin. The primary advance of this work, over the prior reports of “spinal respiratory rhythms”, was that optogenetic methods enabled precise definition of the circuit. Specifically, light activation of Vglut2+ neurons in the cervical cord was sufficient to activate the spinal rhythm. Moreover, activation of this spinal interneuron network could sustain breathing after spinal cord injury, a finding which highlights the importance of understanding and these networks. A similar finding was reported by Satkunendrarajah et al. who showed that stimulation of cervical interneurons could activate the diaphragm after spinal cord injury in mice (Satkunendrarajah et al., 2018).

Our laboratory has occasionally and serendipitously observed possible evidence for spinally derived rhythms in phrenic motor output. The three anecdotal examples in Fig. 3 occurred spontaneously at 2–3 days after incomplete cervical spinal cord injury (C2 hemisection, (Lee et al., 2013; Streeter et al., 2020)). These neurophysiological recordings were made using our previously published methods in mechanically ventilated, vagotomized and urethane anesthetized rats (Fuller et al., 2008; Lee et al., 2013). In these examples, we noted that phasic bursting was occurring in the phrenic nerve ipsilateral to the C2 spinal lesion, but with a distinctly different pattern than was present in the contralateral phrenic nerve. Thus, a novel rhythm is emerging from phrenic motoneurons ipsilateral and caudal to the spinal lesion, and this rhythm appears to be distinct from the primary inspiratory rhythm seen in the contralateral nerve. This pattern was observed to occur during the inspiratory or expiratory phase (Fig. 3A), during exposure to hypoxia (Fig. 3B), and persisted during a period in which lung inflation was not occurring (Fig. 3C). However, no particular intervention was required to evoke these phrenic rhythms – rather these patterns occurred spontaneously during the recordings. These observations are similar to the phrenic rhythms that were recently described in rats following a complete spinal transection (Ghali and Marchenko, 2016), and suggest that spinally generated rhythms (Aoki et al., 1980) can potentially occur simultaneously with the brainstem driven respiratory pattern.

Fig. 3.

Examples of different rhythms in the left versus right phrenic nerve following lateralized C2 spinal cord injury. Recordings are from three different adult male Sprague Dawley rats 2–3 days after spinal injury. A_i) Bursting in the ipsilateral nerve is not synchronous with the contralateral signal. A_ii) The expanded trace, obtained from the box in A_i, shows that ipsilateral bursting can occur during the inspiratory or expiratory (shaded box) period. B) In this example, an apparently non-inspiratory (shaded boxes) rhythm is observed in the ipsilateral nerve. Bii) During brief respiratory stimulation with hypoxia, the respiratory rhythm appears, while the alternate rhythm (shaded boxes) is maintained. C) The ipsilateral rhythmic bursting persists during a period in which mechanical ventilation is removed. This suggests chest wall afferent neurons is not an absolute requirement for rhythmic ipsilateral bursting. TP: tracheal pressure (mmHg) showing the pattern of lung inflation; ABP: arterial blood pressure (mmHg); IL ʃPhr: integrated phrenic neurogram recorded ipsilateral to cervical SCI (a.u.); CL ʃPhr: integrated phrenic neurogram recorded contralateral to cervical SCI (a.u.).

3. Methods for activating the spinal respiratory circuit

Here we review methods for activating the spinal respiratory circuitry. In this context “activation” of spinal circuits refers to utilizing both exogenous (e.g. electrical, chemical, or photostimulation) or endogenous (e.g. via coupling to locomotor circuits during exercise) methods to engage and depolarize spinal neurons which may impact breathing.

3.1. Chemo/Pharmacogenetic

Chemo- or pharmacogenetic activation involves using small molecules, which are not natively recognized by the host. These small molecules are specifically designed to interact with exogenous engineered proteins. One class of chemogenetics are designer receptors exclusively activated by designer drugs (DREADDs) (Roth, 2016). DREADDs allow for targeted activation of neuronal subpopulations (Aldrin-Kirk and Björklund, 2019) through transgenic breeding or viral delivery. Two common receptors, the Gq- and Gi-DREADDs respond to a non-native ligands such as: clozapine-n-oxide (CNO) or Compound 21 (Chen et al., 2015). Gq-DREADD, such as hM3Dq, are muscarinic receptors which increase the excitability of cells expressing them (Armbruster et al., 2007) The Gi-DREADD (hM4Di) results in reduced neuronal excitability. The α7 nicotinic acetylcholine receptor was modified to form another set of engineered channels and ligands known as pharmacologically selective actuator modules (PSAMs) which are activated via pharmacologically selective effectors molecules (PSEMs). PSAMs can then be combined with glycinergic (GlyR) or GABAergic (GABA C) channels to reduce cell membrane resistance and inhibit firing (Atasoy and Sternson, 2018; Magnus et al., 2011). Recent work has suggested that these ligands are not entirely inert (Goutaudier et al., 2019). Indeed, several studies have expressed concerns that CNO can be back metabolized into clozapine, which is biologically active. Of particular concern to respiration, repeated exposure to back metabolized clozapine can blunt CO₂ chemosensitivity (Martinez et al., 2019). While the spinal cord is not the primary site of CO₂ sensing this stresses the importance of DREADD null controls when using CNO or other ligands. Further, chemogenetic approaches are temporally constrained as the cells expressing the novel receptor are tonically excited or inhibited following ligand administration.

A growing number of studies have used chemogenetic methods to manipulate brainstem respiratory circuits, but relatively few have applied this technology towards the study of spinal respiratory interneurons. Satkunendrarajah et al. recently used chemogenetics to manipulate excitatory glutamatergic spinal interneurons following SCI. They injected an adeno-associated virus (AAVDJ) into the murine cervical spinal cord near the phrenic motor pool to produce cre-dependent expression of PSAM^L141F-GlyR in excitatory interneurons. Activation with PSEM³⁰⁸ caused a reduction of inspiratory flow indicating that these spinal interneurons were required to sustain ventilation (Satkunendrarajah et al., 2018). This study also showed that transfection of glutamatergic interneurons in the mid-cervical spinal cord (near the phrenic motor pool) with AAV5-DIO-hM3Dq (i.e., cre-dependent excitatory DREADD), and subsequent activation via CNO causes a substantial increase in diaphragm EMG activity following acute SCI (Satkunendrarajah et al., 2018). A study from Crone’s laboratory used chemogenetic methods to activate V2a interneurons in the mouse (Romer et al., 2017). V2a cells are glutamatergic and are found in the spinal cord and also the brainstem. In the spinal cord, V2a interneurons are synaptically coupled to phrenic motoneurons (Zholudeva et al., 2017). Using a V2a-CHRM3 (Gq excitatory DREADD) mouse, intraperitoneal delivery of the DREADD ligand increased accessory respiratory muscle activity and breathing. These results confirm a role of V2a interneurons in activating the respiratory muscles, but do not shed light on the location of these neurons (i.e., brainstem vs. spinal cord). A subsequent study by Jensen et al. used an inhibitory DREADD to acutely silence brainstem and spinal V2a interneurons in the mouse. The data showed that inhibiting V2a cells caused activation of accessory respiratory muscles, but had minimal impact on the diaphragm (Jensen et al., 2019). Thus, V2a cells may be part of a brainstem and/or spinal cord circuitry which constrains accessory respiratory output during periods of quiet breathing.

Chemogenetic methods are a powerful tool for studying neural networks, and recent publications illustrate that these methods can help to unravel the complexities of the spinal respiratory circuit (Jensen et al., 2019; Romer et al., 2017; Satkunendrarajah et al., 2018). In a relatively short period of time, chemogenetic technology has considerably advanced the understanding of how spinal interneurons can regulate spinal respiratory motor output. As the field moves forward, the use of molecular and genetic tools will enable improved targeting of specific neural populations during chemogenetic stimulation.

3.2. Optogenetic

Optogenetics can selectively target specific cells types, and compared to chemogenetics provides the advantage of high temporal and spatial sensitivity. Since the initial description of activating neurons with light (Boyden et al., 2005) several light sensitive channels have been developed to excite or inhibit neurons at various wavelengths. Transgenic breeding and viral manipulations can be used to insert these channels/pumps into selected cell populations. Transgenic animals tend to have more robust expression, but viral transduction is not as species restrictive. One of the first applications of optogenetics in the spinal cord focused on restoring breathing after SCI (Alilain et al., 2008). In that study, a Sinbus virus was utilized to express channelrhodopsin-2 (Chr2) in the mid-cervical spinal cord of adult rats. Rats received a cervical SCI at the time of viral injection, and several days later, diaphragm EMG activity, recorded ipsilateral to the spinal lesion, could be increased by directing the appropriate wavelength light at the spinal cord. (Alilain et al., 2008). Subsequently, Cregg et al. used optogenetic methods to define a spinal circuit that is capable of rhythmically activating phrenic motoneurons in the absence of brainstem inputs (Cregg et al., 2017). Short (200 ms) pulses of light were used to activate Vglut2+ neurons in the cervical cord, and this was sufficient to generate prolonged (4 s) bursts in the phrenic nerve. However, in the same study, optical activation of the spinal cord in a mouse strain expressing channel rhodopsin in motoneurons (ChAT^Cre; R26R^ChR2) did not produce prolonged bursts. Taken together, this suggests there are spinal glutamatergic neurons capable of generating rhythmic like activity allowing the burst to continue beyond the light duration (Cregg et al., 2017). These studies have shown that optogenetics are a powerful tool for understanding the function of spinal respiratory interneurons, and moreover may be an approach to restore breathing after injuries to the spinal cord.

3.3. Spinal cord electrical stimulation

There is mounting evidence that epidural stimulation (Wagner et al., 2018) or intraspinal microstimulation (ISMS) (Kasten et al., 2013; McPherson et al., 2015) can restore at least partial function in animal models and humans with SCI. For epidural stimulation, electrodes are typically placed directly over the spinal dura; modeling studies indicate that passing electrical currents into the spinal cord with this approach activate sensory afferents and/or ventral spinal roots (Capogrosso et al., 2013). In contrast, ISMS involves placing wires directly into the spinal cord, near the motor neurons. Due to neuronal biophysical properties, ISMS is more likely to activate presynaptic inputs rather than directly depolarizing spinal motoneurons (Nowak and Bullier, 1998a, 1998b). ISMS can also antidromically activate afferents to depolarize spinal motoneurons (Gaunt et al., 2006), and activate multiple motor pools synergistically from one stimulation site (Moritz et al., 2007; Mushahwar et al., 2002). In regards to the focus of this paper, both epidural and intraspinal microstimulation (ISMS) can activate respiratory muscles after SCI, and both methods appear to activate spinal interneurons as summarized next.

The ground breaking work of DiMarco and Kowalski has defined epidural spinal stimulation in the context of respiratory recovery (DiMarco and Kowalski, 2009; DiMarco and Kowalski, 2013a; Dimarco and Kowalski, 2013b; Kowalski et al., 2013). Early work developed a ventral spinal epidural stimulation approach in dogs, and this approach activated all of the primary inspiratory muscles with very few implanted electrodes (DiMarco et al., 1987). These findings led to a clinical intervention combining phrenic nerve and thoracic spinal stimulation (DiMarco et al., 2005). In 2009, this group published a remarkable study that advanced the epidural stimulation approach while also providing new insights regarding spinal respiratory interneurons (DiMarco and Kowalski, 2009). When high frequency (300 pulses per second, pps) stimulation was delivered to the ventral spinal cord of dogs with high cervical spinal transection, it caused recruitment of intercostal and phrenic motoneurons. However, the motoneuron discharge patterns were consistent with “normal” discharge rates (e.g., ~15 Hz). One interpretation of this result is that local spinal circuits were engaged by the high frequency stimulation, and these propriospinal circuits caused “physiologic” recruitment of respiratory motoneurons. Subsequent studies confirmed this fundamental observation (DiMarco and Kowalski, 2013a; Dimarco and Kowalski, 2013b; Kowalski et al., 2017), and also showed that the same approach works in a rat model (Kowalski et al., 2013). Lesioning studies indicate that the upper cervical respiratory interneurons (see Section 2.2) are not involved in recruitment of phrenic motoneurons during high frequency epidural stimulation of the thoracic spinal cord, but that an intact lateral funiculus is required for the response (DiMarco and Kowalski, 2013a; Dimarco and Kowalski, 2013b).

ISMS has also been used to activate spinal respiratory motor output. In a rat model of high cervical SCI (lateral hemisection at C2), the phasic inspiratory signal recorded in the tongue was used as a “trigger” to activate ISMS at C4 (100 pps). This closed loop stimulation effectively activated phrenic motoneurons (indicated by diaphragm EMG recordings) and evoked tracheal pressure fluctuations consistent with inspiratory efforts (Mercier et al., 2017). We also mapped the locations from C2–T1 which ISMS could elicit either independent or simultaneous muscle potentials in the diaphragm, external intercostal, and sternocleidomastoid (Sunshine et al., 2018). Co-activation of inspiratory muscles is more common during inspiration triggered stimulation, suggesting if the timing is appropriate an ISMS based intervention may activate interneuron networks which link these motor pools. The latency of the ISMS-evoked potentials recorded in the respiratory muscles suggested that polysynaptic spinal pathways to respiratory motoneurons were being activated (Sunshine et al., 2018).

Epidural stimulation is already a clinically viable approach to improve respiratory function in individuals with respiratory motor impairment (DiMarco et al., 2009a; DiMarco et al., 2009b; DiMarco et al., 2014), and while not demonstrated in the respiratory circuit in humans ISMS may also be effective in this regard. Strengths of spinal stimulation approaches include the ability to activate respiratory motor pools spanning multiple segments (DiMarco and Kowalski, 2009), evoking “natural” motoneuron recruitment patterns.

3.4. Hypoxia

Reductions in arterial oxygen partial pressure (PaO₂) from typical resting values of 90–100 mmHg triggers a robust activation of carotid chemoafferent neurons which in turn evokes an increase in breathing. Interest in hypoxia has spiked in recent years due to an emerging appreciation that there may be therapeutic value in using carefully controlled mild hypoxia to stimulate neuronal networks after neurologic injury (Gonzalez-Rothi et al., 2015). In regards to this review, the salient point is that hypoxia is a powerful activator of cervical interneurons, and appears to trigger plasticity within cervical networks. Sandhu et al. reported that the majority of mid-cervical spinal interneurons recorded in multi-electrode array experiments changed their discharge pattern during acute exposure to a single bout of hypoxia (inspired O₂ ~ 15%). The data suggested that interneurons responded to hypoxia in a complex but seemingly orchestrated manner, and alterations in neuronal discharge patterns often persisted after the hypoxic stimulus was removed (Sandhu et al., 2015). Subsequently, Streeter et al. reported that inspiratory- and expiratory-modulated cervical interneurons often switched to a tonic firing pattern during exposure to hypoxia. Upon removal of the hypoxic stimulus, cells that displayed inspiratory-related bursting at baseline tended to resume that pattern, but expiratory interneurons maintained the tonic bursting. Interestingly, after repeated exposures to hypoxia (i.e., acute intermittent hypoxia, AIH), cervical interneurons which were activated by hypoxia tended to return to baseline (pre-AIH) firing rates after AIH, whereas inhibited interneurons had persistent decreases in firing rates after AIH (Streeter et al., 2017a). Recent work demonstrated that hypoxia alters the behavior of mid-cervical interneurons in rats with chronic C2Hx (Streeter et al., 2019). At baseline contra- to ipsilesional relationships accounted for ~13% of excitatory connections (assessed via cross correlation), while only ~9% projected ipsi- to contralesional. Following a brief (5 min) hypoxic (inspired O₂ 11%) period this contraipsilesional trend was eliminated and both populations accounted for ~13% of excitatory connections (Streeter et al., 2019).

Thus, hypoxia is a powerful experimental tool for activating (or inhibiting) cervical interneurons, and may have therapeutic value in the context of neurorehabilitation (Gonzalez-Rothi et al., 2015). Mechanistically, it is not known how hypoxia impacts spinal interneuron discharge rates and/or connectivity. These cells may receive synaptic input from brainstem respiratory or sympathetic nuclei.

3.5. Pharmacological agents

Lu et al. observed that thoracic interneurons (T3) responded to application of glutamate at C1–C2. The majority (82%) of the thoracic interneurons responded to glutamate, 55% reduced their firing rate, while 38% increased their firing rate in response to glutamate application (Lu et al., 2004). Phrenic nerve output was largely unaffected by glutamate administration, suggesting there is separate propriospinal network modulating thoracic interneuron firing. To our knowledge, the study by Lu et al. is the only report in which the direct response of respiratory spinal interneurons (e.g., changing in firing rates, biophysical properties) has been evaluated following pharmacologic manipulations. As such, that is an area in need of investigation using extracellular multielectrode arrays (Streeter et al., 2017b) or intracellular recording techniques.

Several studies have induced rhythmic discharge in spinal respiratory nerves (e.g. phrenic, intercostal) following spinal drug application in animal models of high cervical transection. In these experiments, it can be inferred that the rhythmic discharges result, at least in part, from the action of the drugs on spinal interneurons that transmit rhythmic signals to respiratory motoneurons. For example, blocking inhibitory synaptic transmission in the cervical spinal cord can induce rhythmic phrenic/diaphragm discharge after spinal injury. Zimmer used an adult rat C2 hemilesion model, and found that blocking GABA-A, but not glycine receptors in the mid-cervical spinal cord could induce “cross phrenic activity” (Zimmer and Goshgarian, 2007). Ghali and Marchenko subsequently showed that combined application of GABA-A and glycine antagonists applied to the C1–C2 spinal cord could induce rhythmic phrenic bursting in rats that had a high cervical spinal transection (Ghali and Marchenko, 2016). Interestingly, pharmacologic stimulation of the same region of the spinal cord (C1–C2) with glutamate can increase intercostal and phrenic nerve activity (Lu et al., 2004), a finding that indicates synaptic connectivity between C1–C2 neurons and spinal respiratory motoneurons. Finally Cregg et al. have recently confirmed the earlier reports by showing that pharmacologic blockade of inhibitory synaptic transmission in the spinal cord using picrotoxin and strychnine to block inhibitory neurotransmission produced rhythmic bursting in the rat and mouse phrenic nerve following complete transection of the C1 spinal cord (Cregg et al., 2017).

In addition to the aforementioned spinal drug application studies, a few more relevant experiments merit discussion. The early work of Viala and colleagues showed that intravenous delivery of nialamide (monoamine oxidase inhibitor) and DOPA evoked rhythmic phrenic and hindlimb bursting in rabbits with C2 spinal transection. These data provided early evidence for coupling of a spinal locomotor CPG with spinal respiratory circuitry (Viala and Freton, 1983; Viala et al., 1979) (see Section 3.6). Several years later Perségol and Viala used an in vitro brainstem spinal cord preparation from neonatal rats and showed that drugs including 5-HT evoke rhythmic bursting from the cervical ventral roots after transection of the high cervical cord (Perségol and Viala, 1994). Collectively this work suggests cervical neurons are sensitive to chemical stimuli, and these stimuli, under certain conditions, can produce bursts of activity on respiratory nerves.

3.6. Locomotor activity

Activation of lumbar locomotor circuits (Morin and Viala, 2002) as well as limb proprioceptive and/or nociceptive afferents (Amann et al., 2010) can activate respiratory networks. In turn, this may serve to enable (or facilitate) coupling of the locomotor and respiratory rhythms. It has long been known that these two fundamental biological rhythms can be “coupled” in mammals such that breathing movements are coordinated with locomotion (Bramble and Carrier, 1983). The most direct evidence for neurologic mechanisms causing respiratory-locomotor coupling comes from in vitro studies under carefully controlled conditions. For example, electrical activation of hindlimb sensory afferent neurons in a neonatal rat brainstem-spinal cord preparation results in a precise 1:1 entrainment of respiratory and locomotor rhythms (Le Gal et al., 2014). Further, hindlimb flexion-extension movements lead to entrainment of phrenic motor output with limb movement in neonatal rat preparations (Giraudin et al., 2012). Interestingly, stimulation of dorsal roots at cervical and lumbar levels (e.g., where limb muscle afferents are prominent) is much more effective than thoracic dorsal root stimulation at promoting coupling (Giraudin et al., 2012). Thus coupling is likely to involve a neural mechanism triggered via activation of proprioceptive afferents. In regards to spinal interneurons, direct recordings in the in vitro preparation show that expiratory spinal interneurons are also activated in phase with locomotor rhythms. This coupling between spinal respiratory interneuron discharge and locomotor rhythms did not require brainstem synaptic inputs, possibly indicating that the locomotor central pattern generator (CPG) influences spinal respiratory neurons via propriospinal pathways (Le Gal et al., 2016; Le Gal et al., 2014).

Respiratory-locomotor coupling can also be readily observed in exercising humans by examining the timing of each breath relative to each locomotion cycle (Bernasconi and Kohl, 1993; Hill et al., 1988). The ratio of steps to breaths in a given period is typically used to quantify respiratory-locomotor coupling (Daley et al., 2013; Hill et al., 1988; O’Halloran et al., 2012). An example of respiratory-locomotor coupling in humans collected by our research team is shown in Fig. 4. These data were collected in a 41 year old male with a SCI at C5 and classified as a “D” using the American Spinal Cord Injury Association (ASIA) scale. This means more than 50% of the muscles below the level of injury can move against gravity. Note the coupling of the respiratory and locomotor rhythms as indicated by alternating dark and light gray boxes, the number of steps (s) and breaths (b) during each period are marked. This individual with spinal cord injury utilized a 2:1 coupling utilized a 5:2 step to breath ratio, while the spinal injured subject utilized a 2:1 coupling. These data confirm that respiratory locomotor coupling can be observed in persons with SCI as previously reported (Sherman et al., 2009). Further, it has been suggested that a 2:1 entrainment pattern may provide mechanical advantages as the “step driven flow” can reduce the work of respiratory muscles (Daley et al., 2013). In turn, this suggests that spinal neural circuits may underlie the entrainment preference since spinobulbar (and bulbospinal) tracts will be altered after SCI. That suggestion is supported by in vitro work showing entrainment after high cervical transection (Le Gal et al., 2016; Le Gal et al., 2014), but on the other hand brainstem nuclei are involved in coupling when the spinal cord is intact (Giraudin et al., 2012). The mechanisms underlying coupling in the injured human spinal cord are unknown, but one intriguing possibility is that enhancing the strength of respiratory-locomotor coupling (e.g., via neurorehabilitation strategies) could improve breathing ability during exercise. Possibilities such as this are one of the reasons to aggressively pursue a deeper understanding of the spinal respiratory neural circuitry.

Fig. 4.

Examples of respiratory-locomotor coupling. Recordings of knee angle measured with a goniometer and respiration measured using a nasal temperature probe from individuals walking on a treadmill at a self-selected speed. The largest trough in knee angle demarcates a step (s) cycle. The peaks in the respiratory trace indicate a breath (b) cycle. Coupling periods are marked with alternating dark and light gray boxes and the steps (s) and breaths (b) within that period are annotated using the method described previously (Hill et al., 1988; O’Halloran et al., 2012). An example record from a 41 year old male with a C5 motor-incomplete SCI, showing a locomotor respiratory coupling in a pattern of two steps (s1–s2) per one breath (b1).

3.7. Respiratory muscle afferents

Afferent information arising from the respiratory muscles can impact spinal respiratory interneurons and rhythmic respiratory motor output. A few studies suggest that afferents may be a sufficient stimulus to generate rhythmic bursting in respiratory muscles and/or nerves without brainstem input. An early report from Ramos and Mendoza included an example from a spinally transected (C2) rabbit where rhythmic bursting was observed in the diaphragm and intercostals muscles as a result of thoracic distention (Ramos and Mendoza, 1959). A 1977 report from Coglianese et al. showed that after C1–C2 spinal transection in dogs, rhythmic phrenic nerve bursting was observed in sync with positive pressure mechanical ventilation (Coglianese et al., 1977). Since the phrenic nerve bursts occurred in phase with inflation of the lung, it suggests that activation of afferent neurons sensing the movement of the lung and/or chest wall was triggering the response.

A few published studies have provided direct evidence that respiratory muscle afferents can activate spinal interneurons. Bellingham and Lipski found that C5 interneurons in cats could be activated or inhibited when the phrenic nerve was electrically stimulated (Bellingham and Lipski, 1990). Similar data have been reported by other research teams in cats (Iscoe and Duffin, 1996) and guinea pigs (Cleland and Getting, 1993), thus demonstrating that this is a consistent response across species. Razook et al. studied rats and showed that stimulation of putative group III/IV diaphragm afferents (based on conduction velocities) increased bursting in C1–C2 interneurons (Razook et al., 1995). In rats, electrical stimulation of phrenic nerve afferents results in c-Fos activation in mid-cervical spinal interneurons (Malakhova and Davenport, 2001), and anatomical data suggest that myelinated phrenic afferents terminate in the vicinity of pre-phrenic cervical interneurons (Nair et al., 2017a). Finally, activation of propriospinal pathways via phrenic afferent neurons is indicated by studies of “phrenic-to-phrenic” or “phrenic-to-intercostal” reflexes (reviewed in (Nair et al., 2017b)). Thus, it is very well established that activation respiratory muscle afferents can impact the discharge of spinal interneurons. The relative impact of group 1a/1b vs. group III/IV afferents in altering spinal interneuron discharge is not clear, but there is evidence that both proprioceptive and nociceptive afferents can impact spinal respiratory interneurons. It seems likely that one functional impact of these pathways is to enable coordinated activation between the respiratory muscles and/or coordinated activation of respiratory and locomotor muscles.

4. Conclusion – the importance of the spinal respiratory circuit in health and disease

Respiratory-related discharge of spinal interneurons is a common observation across many published studies. The evidence is strong that spinal interneurons can play a role in respiratory motor control in both health and disease, and in some circumstances are capable of driving rhythmic motor discharge in spinal respiratory motor neuron pools. Based on the data reviewed in this article, we suggest that primary functions of propriospinal respiratory neurons include 1) shaping the respiratory pattern into the final efferent motor output from the spinal respiratory nerves (Fig. 5); 2) coordinating respiratory muscle activation across the spinal neuraxis; 3) coordinating postural, locomotor and respiratory movements, and 4) enabling plasticity of respiratory motor output in health and disease.

Fig. 5.

Conceptual diagram illustrating how spinal circuits can shape the pattern of the final motor output. A) The respiratory rhythm (pink) is generated in the brainstem, (some modifications of pattern occur at the level of the brainstem) this command signal is propagated to spinal motor neurons. Interneurons (blue), further shape the final motor pattern (purple). B_i) Following unilateral injury, ipsilateral bulbospinal projections are damaged. Contralateral bulbospinal projections, via propriospinal interneuron, partially compensate. B_ii) Spinal interneuron (green) is activated via electrical stimulation to improve respiratory motor output. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

The potential importance of spinal respiratory interneurons in cases of spinal cord injury merits particular emphasis (Cregg et al., 2017; Hormigo et al., 2017; Satkunendrarajah et al., 2018; Zaki Ghali et al., 2019; Zholudeva et al., 2018). Currently there are few alternatives to mechanical ventilation when independent breathing is not possible. Direct diaphragm pacing can be effective (Posluszny et al., 2014; Smith et al., 2016), but engagement of the local spinal circuitry to activate breathing through electrical stimulation (DiMarco et al., 2018; Kowalski et al., 2013; Sunshine et al., 2018), opto- (Alilain et al., 2008; Cregg et al., 2017) or chemogenetic methods (Jensen et al., 2019; Satkunendrarajah et al., 2018)) may prove to be better options. Spinal interneuron activation with these approaches could allow for synergistic activation of multiple respiratory motor neuron pools. It may also be possible to harness the rhythm generating capacity of spinal networks, either directly (Aoki et al., 1980; Aoki et al., 1978), or via links to locomotor CPGs (Viala and Freton, 1983), to help restore breathing after high cervical SCI.